This method is equivalent to do a serie of narrow up to the first 4 dimensions. It returns a new Tensor which is a sub-tensor going from index dimis to dimie in the =i=-th dimension. Negative values are interpreted index starting from the end: -1 is the last index, -2 is the index before the last index, ... 

> x = torch.Tensor(5, 6):zero()
> print(x)

0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
[torch.Tensor of dimension 5x6]

> y = x:sub(2,4):fill(1) -- y is sub-tensor of x: 
> print(y)               -- dimension 1 starts at index 2, ends at index 4

 1  1  1  1  1  1
 1  1  1  1  1  1
 1  1  1  1  1  1
[torch.Tensor of dimension 3x6]

> print(x)               -- x has been modified!

 0  0  0  0  0  0
 1  1  1  1  1  1
 1  1  1  1  1  1
 1  1  1  1  1  1
 0  0  0  0  0  0
[torch.Tensor of dimension 5x6]

> z = x:sub(2,4,3,4):fill(2) -- we now take a new sub-tensor
> print(z)                   -- dimension 1 starts at index 2, ends at index 4
                             -- dimension 2 starts at index 3, ends at index 4
 2  2
 2  2
 2  2
[torch.Tensor of dimension 3x2]

> print(x)                  -- x has been modified

 0  0  0  0  0  0
 1  1  2  2  1  1
 1  1  2  2  1  1
 1  1  2  2  1  1
 0  0  0  0  0  0
[torch.Tensor of dimension 5x6]

> print(y:sub(-1, -1, 3, 4)) -- negative values = bounds

 2  2
[torch.Tensor of dimension 1x2]